Related Experiment Video
Updated: Sep 26, 2026
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
A scoring model using conventional ultrasound and contrast-enhanced ultrasound to identify benign ACR TI-RADS 4 and 5
Hongyan Song1, Fang Nie2, Tingkun Hou1
1The Second Clinical Medical School, Lanzhou University, Lanzhou, China; Ultrasound Medical Center, The Second Hospital&Clinical Medical School, Lanzhou University, Lanzhou, China.
Background:
The limited specificity of ACR TI-RADS in high-risk categories 4 and 5 results in a substantial proportion of benign nodules being misclassified as suspicious, contributing to unnecessary surgeries. This study aimed to develop and validate a scoring model based on conventional ultrasound (US) and contrast-enhanced ultrasound (CEUS) to better identify benign nodules that may be spared from thyroidectomy.
Methods:
A total of 380 patients with surgically confirmed thyroid nodules (152 benign, 40.0%; 228 malignant, 60.0%) were retrospectively enrolled (January 2021-February 2026). Using stratified random sampling (pathological result, ACR TI-RADS category, background thyroid status), patients were divided into a training set (n = 259) and a validation set (n = 121) at a 7:3 ratio. Independent predictors were identified by multivariate logistic regression, and a scoring model was constructed. Model performance was evaluated by receiver operating characteristic curves, and the area under the curve (AUC), sensitivity, specificity, and accuracy were compared with the ACR TI-RADS using the TR5 cutoff. All validation was performed as internal split-sample validation from a single center.
Results:
In the validation set, the US+CEUS score achieved an AUC of 0.872, sensitivity of 94.4%, and specificity of 75.5%, showing higher performance than ACR TI-RADS (AUC 0.720, sensitivity 88.9%, specificity 55.1%). The US-only score (AUC 0.861, sensitivity 91.7%, specificity 67.3%) served as an alternative when CEUS was unavailable. Subgroup analysis showed that the US+CEUS score achieved a specificity of 92.6% in TR4 nodules and a sensitivity of 96.9% in TR5 nodules.
Conclusion:
The US+CEUS scoring model showed good diagnostic performance in this single-center internal validation cohort of ACR TI-RADS 4/5 nodules, suggesting its potential to reduce unnecessary surgeries by identifying benign nodules that may be spared from thyroidectomy. External validation in prospective, unselected populations is warranted before clinical implementation. The US-only score provides a practical alternative when CEUS is not available.
